Add ClawMoat — host-level runtime security for AI agents

[darfaz]

Project

• Name: ClawMoat
• Repository: github.com/darfaz/clawmoat

Description

Open-source host-level security for AI agents (MIT licensed):

• Host Guardian — 4 permission tiers (observer/worker/standard/full) enforced at runtime
• Forbidden zones — Auto-protects SSH keys, AWS creds, GPG keys, browser data, crypto wallets (20+ patterns)
• Credential monitoring — Watches sensitive directories for unauthorized access
• Skill integrity checking — Hash-based verification + 14 suspicious pattern detectors
• Network egress logging — URL extraction, domain allow/blocklists, 26 blocked domains
• Prompt injection scanning — Multi-layer detection (regex + heuristics)
• Inter-agent message scanning — 10 agent-specific attack patterns
• Policy engine — YAML-based rules for shell, file, browser, network
• Audit trail — Full logging + alert delivery (webhook/file/console)

Key Differentiator

Most AI security tools focus on the prompt/model layer. ClawMoat is the only open-source tool focused on host-level protection — what the agent can actually do on your machine.

Technical Details

• Zero dependencies (pure Node.js)
• Sub-millisecond scanning
• 142 tests passing
• OWASP Top 10 Agentic AI coverage
• Works with any agent framework (LangChain, CrewAI, AutoGen, OpenClaw)

Suggested Section

Runtime Security or Agent Security

npm install -g clawmoat

Website: https://clawmoat.com